While density-functional calculations of the energy are now feasible for biomolecules, the use of density-functional geometry optimizers is still confined to relatively small molecules containing no more than thirty atoms. The key limitation of conventional density-functional geometry optimizers is that the cost of the geometry optimization scales at least quadratically with the number of atoms in the molecule. In contrast the energy at a fixed geometry can be evaluated for a cost which scales linearly with molecule size, enabling very large molecules to be treated. This proposal is based on a radical change in the algorithm for density-functional geometry optimization, potentially reducing the total cost from quadratic to linear in molecule size and enabling a quantum leap in the size of molecules that can be optimized. The proposed algorithm resembles a conventional self-consistent calculation of the energy at a fixed geometry but at convergence the proposed algorithm yields not only the density but also the optimized geometry. This is achieved by simultaneous optimization of the wavefunction and the geometry via a modified self-consistent-field procedure. The proposed algorithm will be implemented in the QChem software package and, if successful, widely distributed through QChem Inc. and Spartan Inc. [unreadable] [unreadable]